Intro to Machine Learning with Google
We already know about online sites that have recommendation engines. Those are one form of personalization tools that use Artificial Intelligence. Another one is voice activated intelligence assistants like Siri but what exactly is machine learning and what can it be used for?
Melody Ivory, a product manager at Google Home, works with Machine Learning every day blending it with hardware and software. In a recent event, she gave us an introduction to Machine Learning.
Product Manager on the Hardware Goods team at Google Home. Before Google, she worked at GE Power’s digital transformation team as a PM for software to control power plants. She has experience in other fields than PM because she has also held PM, UX, and software developer roles at Microsoft, NetRaker, and the Internet Factory and served as a UW professor. Holds a Ph.D./MS in computer science, BS in CS/Math, and MBA.
Distinguishing Machine Learning from other fields
Machine learning, deep learning, A.I… So many different but oh so similar terms. How are they different from each other? Machine learning & algorithms can be used, for example, to define whether a web design is good or for optimizing the way a gas turbine works in a power plant. It can also be used in content suggestion like in Google search. “Machine learning is all about how you teach a computer to learn.” It’s not the same thing as A.I., but it’s a part of it.
Before diving into Machine Learning let’s look at this list we put together to define what Machine Learning and the others really mean.
Computer Science: Theories, experiments, and engineering to inform computer design or use.
Data Science: Methods, processes, and systems to extract insights from data.
Analytics: Discovery of meaningful patterns in data.
Artificial Intelligence/A.I.: Intelligence exhibited by machines to mimic a human mind.
Machine Learning: Computers being able to learn without someone having to hand-code each step. Machine learning is a subset of A.I. that consists mainly of algorithms and data.
Deep Learning: Multi-layered algorithms for learning from data. Deep learning is a subset of machine learning, and it’s about doing things with a lot of data and A.I.
Machine Learning and Algorithms
Now let’s dig into it. The main thing in machine learning is that you have to ask good questions to get good answers. Before you can even start the process, you have to know what you want to know. After that, you think about the algorithms that you’re going to use as part of the process. “The algorithms dictate the type of data that you need to collect or how you need to prepare it to go into the algorithms.”
The algorithms determine what kind of output you get, so you also need to know the type of answers you want before you start doing anything. The algorithms have to be determined before collecting the data. The key is to use good data to get good data. After the process is finished, it’ll be repeated with an improved model.
Important questions used in Machine Learning
Like mentioned before the most important thing in Machine Learning is asking good questions. There are different questions in separate stages of the process that can help answer these various questions well.
The question what is to monitor. If we do this then what. Why diagnoses the situation. When we did this why did that happen? Why adds more value and more information that you can act on but both of these questions answer to something that has already happened in the past and therefore they only give you information.
Predicting question when can tell you what will happen and it adds even more value than why because it can give insight at the moment. The last stage is optimizing with questions like what if and how to. They can tell you in advance when something is going happen and give you a chance to do something about it. Questions can drive value to Machine Learning.
Machine Learning process
The process itself is divided into two phases; learning and predicting. Your goal is to get a model done by using algorithms. In the first phase you start by choosing the questions you want to ask and the algorithms you want to use. Next, you collect the data and clean it. This takes most of the time because it can take a while getting the good data and cleaning it up.
The next step you build and evaluate the models and algorithms, and lastly, you deploy the model. In the second phase, you put your new data into the deployed model to get the answers you’ve wanted all along, and the last step is adapting.
Questions from the audience
How did you get the algorithms to define whether your web design was good?
Keep in mind that I did that in 2001, but I used algorithms from IBM SPSS which is a statistics program. Today I would use R Project because it has all these algorithms for free. It also has different program environments to do, for example, commandment lines and batch stuff. The reality is that you don’t know exactly what parameters to use. Every algorithm has a different sense and a lot of the times you just have to try it over and over again to see the results.
Can you talk about a product that was supposed to be using machine learning that did not work?
I have not worked on a product that tried to use machine learning and didn’t work because if it doesn’t work you used the wrong algorithms or you got the wrong data. You just need to keep trying until it does. Sometimes you have to make different transformations to your data to get it to work.
When you step into the hat of “I’m building something in machine learning” you don’t stop until you’ve built something that works and you through everything at it. That’s how it works.
What are the opportunities working in this field part-time?
You can get experience and get paid through, for instance, Kaggle, a crowd source data mining company, where you do competitions building models. Whoever builds the best model wins prices. That’s one way to do it part-time and on your own time. You choose which things you want to compete in. If you win, you get money.
But if you say I want to get paid for this 30 hours that I worked I’d say consulting is a better opportunity. On Upwork the fastest growing category is machine learning, people with machine learning expertise. If you have that expertise, there are a lot of startups and a lot of larger companies looking to bring freelancers into the company, so there are easy ways to get into it.
Machine learning is on top of the world right now. It’s something that everybody’s talking about and that a lot of companies are investing in. It provides, even more, opportunities for Product Managers and more interesting things to gain knowledge about, and on top of everything, it’s fascinating! Get into it and build your skillset!
Have any comments? Tweet us @ProductSchool
We teach product management courses in San Francisco, Santa Clara, Los Angeles, and New York. To learn more about our upcoming courses and how to apply click over to our course page.